A General Approach of Automated Environment Design for Learning the Optimal Power Flow
Thomas Wolgast, Astrid Nieße
TL;DR
The paper introduces a general, automated RL environment design framework for learning the Optimal Power Flow (OPF) using multi-objective hyperparameter optimization. By treating environment choices (data, observations, rewards, episodes, actions) as tunable hyperparameters, the method builds inner RL training runs and an outer HPO loop to optimize both constraint satisfaction and objective minimization. Across five OPF benchmarks, the automated designs consistently outperform a manually crafted baseline and yield new insights into effective environment designs, though overfitting to a particular RL algorithm remains a risk. The approach is generalizable to other RL domains and highlights the trade-offs and data-design considerations essential for robust RL in power systems.
Abstract
Reinforcement learning (RL) algorithms are increasingly used to solve the optimal power flow (OPF) problem. Yet, the question of how to design RL environments to maximize training performance remains unanswered, both for the OPF and the general case. We propose a general approach for automated RL environment design by utilizing multi-objective optimization. For that, we use the hyperparameter optimization (HPO) framework, which allows the reuse of existing HPO algorithms and methods. On five OPF benchmark problems, we demonstrate that our automated design approach consistently outperforms a manually created baseline environment design. Further, we use statistical analyses to determine which environment design decisions are especially important for performance, resulting in multiple novel insights on how RL-OPF environments should be designed. Finally, we discuss the risk of overfitting the environment to the utilized RL algorithm. To the best of our knowledge, this is the first general approach for automated RL environment design.